Deft is an e-commerce search engine that utilizes knowledge graphs and language models. Pulse 2.0 interviewed Deft CEO and co-founder Zach Hudson and CTO and co-founder Alex Gunnarson to learn more.
(left, Alex Gunnarson, co-founder and CTO, right, Zach Hudson, co-founder and CEO)
Background Of The Founders
Could you tell me more about your background?
Gunnarson said: “I have ten years of experience in engineering in startup contexts, including a background in linguistics and formal logic (degrees). I was the first hire for what became Roam Research, and I have been thinking about the problem of transforming unstructured data into knowledge graphs — i.e., parsing the web — ever since then. I wrote a database from scratch that outperforms RedisGraph (state-of-the-art) by 700x on product queries.”
Hudson said: “I was building Rcmmd.com, which was ‘Yelp for product reviews’ before joining Deft. Prior to that, I worked as a Director of Product for major e-commerce brands and technologies for 11 years. The problem of trust and authenticity in e-commerce has been a challenge I’ve been working on for half a decade. I taught myself programming in order to build Rcmmd.”
“We’re both Pioneer winners, On Deck Fellows, and part of Jason Calacanis’s LAUNCH.”
Formation Of Deft
How did the idea for Deft come together, and what are your primary responsibilities? Hudson shared:
“The idea for Deft started back in 2019 when our cofounder Alex was in the market for a gray, mid-century couch with wood trim on the bottom for under $800. It took him 30 hours to find it, scrolling through various sites and opening hundreds of tabs. He realized others had similar experiences and started building something minimal to meet their needs. Alex was building the early days of Deft, and I was moonlighting on my project, Rcmmd. We bumped into each other at a startup event and kept in touch since we were working in a similar space and became friends.”
“Shortly after, we started doing work for each other’s projects. For about four months, I was assisting Alex with strategy and growth for Deft. Alex was helping me select ML and DB models for better review sentiment analysis.”
“Eventually, we realized we were working on the same problem but from different angles. I shut down Rcmmd and joined Alex on Deft. We won Pioneer, and then we were accepted into the Launch accelerator after we got some early traction. A joint study conducted by Google and Inmar found that the average shopper spends 15 hours searching across 12 websites over the course of 79 days before making a purchase decision, and it’s only getting worse. The introduction of large language models (LLMs) across years of over-SEO’d data is a dangerous cocktail. With it, the line between truth and fiction becomes more fuzzy. With over 2 billion products from reputable shops and growing, how do you find the right one?
Favorite Memory
What has been your favorite memory working for the company so far? Hudson reflected:
“Making our first users happy and getting their emails is certainly one of my favorites, but let me give you a shared experience from Alex and me. There have been some long nights that resulted in technical breakthroughs. Here’s one of the more instrumental stories: We were in desperate need of a way to handle the scale of our product index and the concurrent searches happening from our users. Alex hatched an idea to create a new type of database to get around the problems we were seeing with our current one. It was a huge risk and uncharted territory. After a few months of working on it, Alex finally had a breakthrough that would make the new DB possible. The day after the final engineering marathon, we were running the numbers to compare our new database vs. the state-of-the-art RedisGraph. Our new database is 700x faster. We were ecstatic and definitely did not expect such a monumental outcome. Those marathons paid off.
What are the company’s core products and features?
– Natural language search People can’t communicate their vision for a product in a way that e-commerce sites understand, so they scroll through 100s of pages of ad-riddled results. They try to communicate by typing, but Amazon and almost every other product search engine uses keyword search. This is fine for general searches (‘gray couch’) or make-and-model ones (‘Article Anton Gray Sofa’) but breaks down for most others. Deft fixes this by offering a natural language search.
– Image Search Power users sometimes try to communicate their vision by image. They’ll try a Google image search, but they can’t restrict this search to only products, so they get irrelevant results. Pinterest image search helps, but results aren’t restricted to products either. Deft’s image search fixes this. It only searches over products, and though users can upload images, they don’t need to have one handy — they can just select a product image already in their search results. What if you like the look of a product, but you want it in a different color or price? With Deft, you can combine text and images to find exactly what you’re looking for. For example: Just upload a photo and say “like this, but under $400”.
– Ad-free and SEO-free Deft is ad-free because we conjecture that ads yield short-term gain but long-term customer erosion. As Larry and Sergey say, ‘advertising funded search engines will be inherently biased towards the advertisers and away from the needs of the consumers.’ SEO is fundamentally broken. There are whole categories of searches — in e-commerce especially — that have been so ‘over-SEO’d’ that searching them on Google or Amazon is essentially useless. Paul Graham echoes our sentiments. Our system can’t be manipulated by SEO because it relies on verified product knowledge.
– Aggregate from around the web. We’re still in the early days of this, but we aggregate data from around the web so you can make your decision more quickly. Based on your query, we pull in relevant reviews, product photos, and more.
– Upcoming Features: Browser extension for Deft, AI Shopping Concierge, and new categories, such as clothing/apparel, electronics, and more. We envision our browser extension to be similar to Honey but for search. Users will never have to leave the store or site to get high-quality search results. The AI Concierge allows you to have a conversation with our shopping assistant, who can recommend products like you were talking to a real expert in the stores.”
Challenges Faced
Have you faced any specific bottlenecks in your sector of work recently? Hudson acknowledged:
“Interestingly enough, the biggest challenge in search isn’t the search algorithm; it’s cleaning up the bad data and parsing it. We’re at the point where we can ingest millions of products a month, but that’s still too slow considering there are over 2 billion products across reputable sites in the US alone and growing.”
Evolution Of Deft’s Technology
How has the company’s technology evolved since launching? Hudson noted:
“The first version of Deft was an online form that, when you submitted it, would be sent to Alex and me. We will manually compile the results and email them back to you within 72 hours. We wanted to get a sense of what types of queries people were going to do. The next version was our closed Beta. It was very limited in its ability to understand natural language, and we also had a very small number of products available to search. The latest version is where we’ve made the most strides. We’ve launched publicly with our natural language and multimodal search. You can search across hundreds of thousands of products instantly.”
Significant Milestones
What have been some of the company’s most significant milestones? Hudson cited:
“In the beginning, it was your standard startup milestones: The first version of Deft. The first query on Deft. The first user of Deft. The first non-co-founding engineer at Deft. Here are some more recent ones: – Deft is now one of the largest — if not the largest — catalogs of home decor and furniture on the internet. – Our knowledge graph makes accurate natural language queries possible and is 700x faster than state-of-the-art technologies.”
Customer Success Stories
After asking Hudson about customer success stories, he cited:
“One of the very first users of Deft had been looking for a couch for about six months. We found her on Reddit, where she had made a post essentially giving up on her hunt for the perfect couch. We asked her to try Deft. She found the product in a single day. She immediately came back and furnished her entire living room using Deft’s search. She continues to use us on a monthly basis. Here are some other fun quotes from users:
– “That was AMAZING! You saved me SOOOO much time and frustration finding that darn thing.”
– “‘So far, I’m super impressed with the recommendations and level of detail. Honestly never would have been able to find these.”
– “This is wonderful! How did Deft even find this? My husband took a photo and tried all sorts of image searches but came up blank.”
Funding/Revenue
After asking the Deft founders about funding and revenue information, Hudson revealed:
“1.) Funding: We’ve raised $1.7mm in a pre-seed round from Frontier VC, Hustle Fund, VITALIZE, Long Ecommerce Ventures, 43 Ventures, LAUNCH, and over 15 founders and execs from industry leaders like Apple, Stripe, Affirm, and Honey.
2.) In terms of revenue, we’ve been testing our subscription and usage-based models, but it’s nothing we’re ready to share. What’s important here is that the business model of search needs to change. Advertising aligns incentives with brands, not consumers. You can’t have the most accurate search when you allow placement to be bought and sold to the highest bidder as Google and Amazon do. At Deft, we’re focused on finding the right model that aligns us with consumers (our users).”
Total Addressable Market
What total addressable market (TAM) size is the company pursuing? Hudson assessed:
“Deft has an enormous market opportunity. Even if we only stayed in home decor and furniture, we would see a TAM mirroring Wayfair’s, which reported $12 billion in revenue last year and is one of the 250 largest companies in the country. We plan to expand across categories and tackle e-commerce search at large which easily presents a TAM of >$800B. Think Amazon 2.0, but without having to deal with warehousing/fulfillment complexities.”
“According to Statista, 290 million people shopped or browsed products online just in the US (source). Worldwide, it grew by 1 billion between 2019 and 2022 and is set to keep growing.”
Differentiation From The Competition
What differentiates the company from its competition? Hudson assessed:
“We have a proprietary hybrid search engine that doesn’t follow the trend of exclusively deep-learning approaches like GPT. We also have one of the cleanest e-commerce data sets in existence. We’ll continue to grow it, train our models, build our knowledge graph with it, and ultimately use it to give our users the best shopping experience on the internet. When we think of competitors, Amazon is the biggest and most obvious player in the e-commerce space. Google Shopping is another big one, but its consumer adoption is low and its conversion rate is reportedly low. Both of them are dependent on an ad-based, SEO-gameable model, which decreases relevance and erodes trust. The intrinsic differentiator of Deft is trust. We’re aligning ourselves with the customer instead of the brand. Trust is paramount when spending money online. It’s incredibly hard to build and very easy to lose. This is a search engine built for you, not to sell you junk.”
Future Company Goals
What are some of the company’s future company goals? Hudson concluded:
– Our next user milestone is 50k monthly active users (MAU).
– $2 million annual recurring revenue.
Additional Thoughts
Any other topics to discuss? Hudson concluded:
“LLMs have unlocked some of the first innovations we’ve seen in search in years. However, you can’t shortcut your way to better search by adding an “intuition layer” (GPT) on top of bad data, which is the approach a lot of companies are taking. Instead, we’ve assembled our own knowledge graph and reasoning engine to harness the power of models like GPT while taming their pitfalls. Respected researchers like Meta’s Chief AI scientist agree that an approach like this is needed to enable the next intelligence leap for AI.”
